Hands, Fingers, and the Classic Among AI Errors
When you check an image, your first look should almost always go to the hands. For a long time, they were the most reliable weak spot of image generators. Pay attention to the number of fingers, because AI models sometimes produce six fingers, leave one out, or merge two into one. The shape can be a giveaway too: fingers may look bent, of uneven length, kinked in the wrong direction, or ending in an unnatural point.
On top of that, hands often do not match the rest of the scene. A hand holding an object sometimes grasps at empty space or blends into the object. When two people shake hands, the fingers can merge into each other. Feet, toes, and limbs are worth a close look as well, because the same problems show up there. Important: newer models have caught up significantly on hands. Correct hands are no longer proof of a real photo today.
The Face in Detail: Eyes, Teeth, Ears, and Symmetry
Faces often look perfect at first glance, but in the details something is frequently off. Look closely at the eyes: the two pupils can be different in size or shape, and the light reflections sometimes do not match, even though both eyes should be seeing the same light source. Teeth are another weak spot. They may look too numerous, merge into a single row, or have an odd size and arrangement.
Ears deserve attention too, because their complicated structure is hard for AI models to get right. Often a person's two ears look completely different or have an unclear inner shape. In general, it helps to check for symmetry: glasses with two differently shaped lenses, mismatched earrings, or a face whose left and right halves do not quite fit together are typical warning signs.
A final detail in the face is the skin. AI images often show a surface that is too smooth, almost waxy, with no pores, no fine wrinkles, and no small irregularities. If a portrait looks like an advertisement even though it depicts an everyday situation, caution is warranted. That said, real photos also use strong beauty filters, so that alone is no proof.
Text, Background, and Reflections
Text in an image is one of the toughest tests for AI. Look at signs, logos, book titles, lettering on clothing, or house numbers. You will often spot letters that do not form any meaningful text, distorted or invented characters, uneven spacing, or a word that looks readable from a distance and falls apart when you zoom in. Real photos usually show clean, legible text here.
The background reveals a lot too. Watch for illogical transitions, for lines behind a person that do not line up, for railings or window frames that suddenly bend, and for objects that trail off into nothing. Reflections are a chapter of their own: a mirror showing something different from the scene in front of it, a shop window with a mismatched reflection, or a water surface that reproduces the subject incorrectly all point toward AI.
Lighting and shadows should also fit together. If the light comes from the left but the shadow also falls to the left, or if several people cast contradictory shadows, something is wrong. Such physical inconsistencies are often easier for the eye to grasp than the finer detail errors.
Patterns, Accessories, and Merging Objects
AI models struggle with regular, repeating structures. Watch out for tiles, fences, blinds, keyboards, checkered patterns, or crowds in the background. These patterns often break off in the middle, warp, or repeat in unnatural ways. In a crowd, you will also frequently find faces that are blurry, distorted, or obviously wrong.
Jewelry and accessories are another weak spot. Necklaces that blend into the skin, earrings that merge with the hair, temple arms of glasses that disappear into the head, or watch straps with no clear end are all typical. The same applies to the transition between clothing and body, and to straps, buttons, and zippers. If a detail is not clearly separated when you look closely but instead blends into its surroundings, that is a clear indication.
Why These Errors Happen and Why They Disappear
Most AI images today come from what are called diffusion models. Put simply, these models learn to work a coherent image out of random image noise, step by step. In doing so, they produce what looks statistically plausible, without any real understanding of anatomy, physics, or language. That is why errors appear exactly where precise logic matters: in the exact number of fingers, in legible text, in correct shadows, or in the question of what a mirror should show.
It is important to understand that these weaknesses are not a fixed state. With every new model generation, signs that used to be reliable disappear. Hands with the right number of fingers, clean eyes, and even short pieces of text often come out flawlessly today. That means two things: first, the absence of an error is no proof of a real photo. Second, you should not rely on a single anomaly, but consider several signals together.
Why the Eye Is Not Enough and What Really Helps
As useful as these visual signs are, they have a clear limit. Good AI images today often no longer contain a single visible error, and at the same time real photos can look artificial due to heavy editing, filters, or poor lighting. Anyone who relies on their own eye alone will therefore be wrong regularly, and in both directions. Even trained eyes reach their limits as the models keep getting better.
That is why a combination makes sense. Use the visual check as a first filter and, when in doubt, add a technical, forensic analysis. Such methods do not look at the obvious subject but at traces in the image data, at metadata, and at statistical anomalies that the eye cannot see at all. If you want a specific image or link assessed, you can upload it to us or submit the address and have it checked. A technical analysis does not deliver a final verdict with absolute certainty either, but it adds a layer to your first impression that you cannot reach with the naked eye.